Summary: Wiz’s November investigation into abused Qubole integration roles showed how idle vendor access can become an attacker’s easiest path to sensitive data. This article argues that declining integration usage should trigger a review of the data an identity can still reach, and it gives security and data teams a five-step approach for retiring vendor access with verified, documented evidence.
Most security teams have a list of vendors they no longer think about. The pilot ended, the contract lapsed, or the team that championed the tool moved on to something else. Once an integration stops doing anything useful, it tends to stop getting attention too. The problem is that an integration that has stopped being used and an integration that has stopped having access are two very different things, and attackers understand the gap between them very well.
On November 9, an investigation conducted with Sentra and Mandiant, describing how a threat actor used IAM credentials from an AWS account associated with Qubole to assume integration roles in customer AWS environments between October 14 and October 26. The attacker stole stored secrets, cloud credentials, and sensitive files. What makes the case instructive is not the technique. It was found that most of the abused roles were otherwise unused, and in a majority of the environments it observed, the attacker generated the only role-assumption activity from the vendor account during October. The access was still available because the trust policies that granted it were still in place.
In other words, an idle integration became the way in, and the sensitive data it could reach was the prize.
Idle access is still access
That incident played out in AWS identity and storage, but the pattern is not limited to cloud trust policies. It shows up one layer down as well, inside the data platforms where an organization keeps its most sensitive tables. A BI tool, an ETL service, a marketing platform or an analytics vendor receives a service account during onboarding. It is given grants broad enough to get the project moving. Months later, the project winds down, while the service account quietly keeps its role, its grants and often a working credential.
Usage data can reveal a lot about this kind of exposure, but most teams only read it in one direction. A spike in activity raises an alert. A slow decline rarely raises anything at all. We think it should. When an integration’s usage drops off, that is the right moment to ask a simple question: what sensitive data can this identity still reach, and does anyone still need it to?
Why retiring an integration needs evidence, not just intent
Deciding to retire an integration is easy. Proving that it has actually been retired is harder. Vendor access is usually spread across several systems at once, including a cloud trust relationship, a database role, a set of object grants, a credential stored in a secrets manager and sometimes an active session or token that outlives the credential that created it. Removing one of those pieces can leave the others intact, which means the access path has been narrowed rather than closed.
There is also a risk running in the opposite direction. Pulling permissions without checking dependencies can break a quarterly report or a downstream pipeline that nobody remembered was connected. That is why business owners belong in the process. They can confirm what the integration still does, if anything, before security removes its access. Done well, retirement is a short, coordinated exercise with a clear end state and a record showing that it happened.
A practical approach to reviewing vendor access to sensitive data
The five steps below give security, data and business teams a shared sequence to follow. They work whether you are cleaning up a single dormant vendor or building a recurring review cycle.
- Inventory vendor-linked identities and their connections. Start on the database side with the service accounts, roles and users created for third-party tools, along with the grants attached to them. Then map each one to the related cloud trust relationships, API keys or stored credentials that your cloud and identity teams manage. The goal is one view of every path a vendor could use to reach your data.
- Identify the sensitive data each identity can reach. Permissions only tell part of the story. Read access to a schema full of reference data is a minor concern, while the same access to a schema holding customer PII or payment details is a serious one. Mapping effective permissions against classified tables and columns shows where the real exposure sits.
- Compare what each identity can do with what it actually does. Query history shows whether an integration still touches the data it was granted. If a vendor account can read twenty sensitive tables but has queried two of them in the last ninety days, or none at all, you have a strong case for narrowing or removing its access. Bring that evidence to the business owner and confirm whether any remaining need exists before you act.
- Coordinate removal across the responsible systems. Revoke the database grants, disable or drop the service account, retire credentials in the secrets manager and revoke active sessions. The teams that own cloud trust policies and IAM roles should remove them on the same timeline. Handling these steps together, rather than leaving pieces for later, closes the gaps an attacker would otherwise use.
- Verify that the retired paths fail and keep the record. Test that the retired credentials and roles no longer work. Then retain evidence of what was removed, when it was removed, who removed it and who approved it. That record matters for audits, and it matters just as much when the next incident report lands and leadership asks whether your organization is exposed.
Where ALTR fits
ALTR works at the data layer, which is where the second and third steps become much easier to sustain over time. ALTR classifies sensitive data and shows which roles and users are accessing which sensitive columns, how much data they retrieve and how often. Query audit logs record who accessed what and how, and usage analytics show consumption trends by user and role. Together, those views make it possible to spot an integration whose activity has faded while its reach into sensitive data has not.
ALTR’s policy engine can also narrow what a service account is able to see through masking or tokenization while business owners confirm dependencies, and it can alert security teams or automatically block access when activity falls outside defined norms for time, rate or volume. The cloud trust relationships themselves remain with your cloud and identity teams. ALTR’s role is to make sure the sensitive data those identities can reach is known, governed and monitored, so that retirement decisions rest on evidence rather than assumptions.
Make declining usage a trigger for review
The Qubole investigation is a reminder that trust often outlasts the relationship it was created for. Not every dormant integration will be targeted, but every one of them represents access that delivers no business value while still carrying risk. Treating a decline in usage as a signal, and following it through to a verified and documented retirement, is one of the most practical ways to reduce exposure your organization is not actively using.
A good place to start is with the vendors nobody has mentioned in six months. Those are usually the ones worth checking first.